Application Of Peripheral Blood Cell Classification And Counting In Deep Learning

In the medical laboratory,peripheral blood cell test(blood routine),as a convenient and accurate method for sampling,has always been of great significance for the determination of diseases.Peripheral blood cell detection is mainly used to determine whether a patient has a certain disease by identifying all kinds of cells in the blood and calculating whether the proportion of all kinds of cells is within the normal range.The results of different blood tests reveal the mechanism of several important blood disorders.In the field of medical image processing,with the great progress of imaging technology,using computer graphics auxiliary diagnosis become a big trend,on the one hand,the development of imaging technology has brought the huge amounts of medical data,on the other hand,computer graphics,auxiliary diagnosis of blood sample images can be generated,more accurate,more efficient diagnosis.In short,how to apply the deep neural network to medical detection,classify and count the blood sample images taken by computer with the deep neural network,and replace the manual operation of doctors has become a hot topic of wide attention.In cooperation with Nanjing Gulou Hospital,this project collected microscopic images of clinical peripheral blood cells in the department of medical laboratory and established data sets of peripheral blood cells.This subject USES the method of deep learning to classify and count blood cells.The main work includes the following:1.After analyzing the difficulties in the collection of peripheral blood cell images,we established our own blood cell data set.Starting from the actual situation,this data set has complete classification,4 more than the existing 5 types of white blood cell data set,laying a foundation for the subsequent classification and counting research.2.Aiming at the problem of unbalanced data quantity among data sets,data enhancement and random sampling are carried out for the training set.A new data enhancement method was designed to increase the sampling frequency and improve the test accuracy to 99%.3.According to the research status of erythrocyte count,the accuracy of erythrocyte count is increased to 97.5%by using the counting method based on convolutional neural network and combining the feature pyramid(FPN)and deep residual network ResNet,starting from image processing.This subject has basically completed the classification and counting of blood cells,solved the classification difficulty caused by the excessively high similarity among cells,and covered almost all cell types in peripheral blood that affect the judgment of diseases.This research has greatly saved the manpower cost,and has laid the foundation for the follow-up research.